An Ensemble Learning-Based Approach for CBC Anemia Classification

Abstract

Anemia is one of the most prevalent hematological disorders worldwide, affecting billions and leading to serious health complications. Early detection and accurate classification of anemia types are essential for effective treatment and management. This research proposes a machine learning-based framework for classifying nine distinct types of anemia using Complete Blood Count (CBC) data. The methodology integrates two feature selection techniques Recursive Feature Elimination (RFE) and heatmap correlation analysis to identify the most relevant biomarkers. A stacking ensemble classifier was developed using Random Forest, Support Vector Machine, K-Nearest Neighbors, and Decision Tree as base learners, with logistic regression as the meta-learner. Experimental evaluation showed that the (RFE) derived features yielded superior performance, the Anemia datasets that are available on Kaggle   platform was used of which consists of 1,281 observations. The data was balanced to prevent the model from favoring the dominant class in the data. with the final model achieving an accuracy of 99.83%, and F1-score of 99.83%. The results demonstrate the effectiveness of ensemble learning and advanced feature selection in improving diagnostic accuracy and provide strong evidence for deploying intelligent decision-support systems in clinical settings. The generalizability of the proposed model was rigorously evaluated using 1,200 new, unlabeled CBC samples obtained from local laboratories in Misrata. In a blinded clinical validation research, the model’s predictions were compared against diagnoses made by expert hematologists, demonstrating strong concordance with an overall agreement of 84.67% across all anemia subtypes. These results underscore the effectiveness of ensemble learning combined with advanced feature selection.